Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <cc9e6464-5e47-4044-9785-c57167f0e1c5@linaro.org>
Date:   Wed, 9 Aug 2023 18:52:17 +0200
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH 03/15] clk: qcom: gcc-sm6375: Unregister critical clocks
Content-Language: en-US
To:     Johan Hovold <johan@kernel.org>,
        Bjorn Andersson <andersson@kernel.org>
Cc:     Andy Gross <agross@kernel.org>,
        Michael Turquette <mturquette@baylibre.com>,
        Stephen Boyd <sboyd@kernel.org>,
        Rob Herring <robh+dt@kernel.org>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@linaro.org>,
        Conor Dooley <conor+dt@kernel.org>,
        Marijn Suijten <marijn.suijten@somainline.org>,
        linux-arm-msm@vger.kernel.org, linux-clk@vger.kernel.org,
        linux-kernel@vger.kernel.org, devicetree@vger.kernel.org
References: <20230717-topic-branch_aon_cleanup-v1-0-27784d27a4f4@linaro.org>
 <20230717-topic-branch_aon_cleanup-v1-3-27784d27a4f4@linaro.org>
 <ZLaRtrH85v4kpSvb@hovoldconsulting.com>
 <33a26241-026a-9466-5dd6-e3202b29f57c@linaro.org>
 <ybugl2m7o5cnzj4lv5ksit2rip6yvths5ieo3xlw6cycto2zax@2jimga475z2t>
 <ZLeiM6l6tu6XDzrr@hovoldconsulting.com>
From:   Konrad Dybcio <konrad.dybcio@linaro.org>
Autocrypt: addr=konrad.dybcio@linaro.org; keydata=
 xsFNBF9ALYUBEADWAhxdTBWrwAgDQQzc1O/bJ5O7b6cXYxwbBd9xKP7MICh5YA0DcCjJSOum
 BB/OmIWU6X+LZW6P88ZmHe+KeyABLMP5s1tJNK1j4ntT7mECcWZDzafPWF4F6m4WJOG27kTJ
 HGWdmtO+RvadOVi6CoUDqALsmfS3MUG5Pj2Ne9+0jRg4hEnB92AyF9rW2G3qisFcwPgvatt7
 TXD5E38mLyOPOUyXNj9XpDbt1hNwKQfiidmPh5e7VNAWRnW1iCMMoKqzM1Anzq7e5Afyeifz
 zRcQPLaqrPjnKqZGL2BKQSZDh6NkI5ZLRhhHQf61fkWcUpTp1oDC6jWVfT7hwRVIQLrrNj9G
 MpPzrlN4YuAqKeIer1FMt8cq64ifgTzxHzXsMcUdclzq2LTk2RXaPl6Jg/IXWqUClJHbamSk
 t1bfif3SnmhA6TiNvEpDKPiT3IDs42THU6ygslrBxyROQPWLI9IL1y8S6RtEh8H+NZQWZNzm
 UQ3imZirlPjxZtvz1BtnnBWS06e7x/UEAguj7VHCuymVgpl2Za17d1jj81YN5Rp5L9GXxkV1
 aUEwONM3eCI3qcYm5JNc5X+JthZOWsbIPSC1Rhxz3JmWIwP1udr5E3oNRe9u2LIEq+wH/toH
 kpPDhTeMkvt4KfE5m5ercid9+ZXAqoaYLUL4HCEw+HW0DXcKDwARAQABzShLb25yYWQgRHli
 Y2lvIDxrb25yYWQuZHliY2lvQGxpbmFyby5vcmc+wsGOBBMBCAA4FiEEU24if9oCL2zdAAQV
 R4cBcg5dfFgFAmQ5bqwCGwMFCwkIBwIGFQoJCAsCBBYCAwECHgECF4AACgkQR4cBcg5dfFjO
 BQ//YQV6fkbqQCceYebGg6TiisWCy8LG77zV7DB0VMIWJv7Km7Sz0QQrHQVzhEr3trNenZrf
 yy+o2tQOF2biICzbLM8oyQPY8B///KJTWI2khoB8IJSJq3kNG68NjPg2vkP6CMltC/X3ohAo
 xL2UgwN5vj74QnlNneOjc0vGbtA7zURNhTz5P/YuTudCqcAbxJkbqZM4WymjQhe0XgwHLkiH
 5LHSZ31MRKp/+4Kqs4DTXMctc7vFhtUdmatAExDKw8oEz5NbskKbW+qHjW1XUcUIrxRr667V
 GWH6MkVceT9ZBrtLoSzMLYaQXvi3sSAup0qiJiBYszc/VOu3RbIpNLRcXN3KYuxdQAptacTE
 mA+5+4Y4DfC3rUSun+hWLDeac9z9jjHm5rE998OqZnOU9aztbd6zQG5VL6EKgsVXAZD4D3RP
 x1NaAjdA3MD06eyvbOWiA5NSzIcC8UIQvgx09xm7dThCuQYJR4Yxjd+9JPJHI6apzNZpDGvQ
 BBZzvwxV6L1CojUEpnilmMG1ZOTstktWpNzw3G2Gis0XihDUef0MWVsQYJAl0wfiv/0By+XK
 mm2zRR+l/dnzxnlbgJ5pO0imC2w0TVxLkAp0eo0LHw619finad2u6UPQAkZ4oj++iIGrJkt5
 Lkn2XgB+IW8ESflz6nDY3b5KQRF8Z6XLP0+IEdLOOARkOW7yEgorBgEEAZdVAQUBAQdAwmUx
 xrbSCx2ksDxz7rFFGX1KmTkdRtcgC6F3NfuNYkYDAQgHwsF2BBgBCAAgFiEEU24if9oCL2zd
 AAQVR4cBcg5dfFgFAmQ5bvICGwwACgkQR4cBcg5dfFju1Q//Xta1ShwL0MLSC1KL1lXGXeRM
 8arzfyiB5wJ9tb9U/nZvhhdfilEDLe0jKJY0RJErbdRHsalwQCrtq/1ewQpMpsRxXzAjgfRN
 jc4tgxRWmI+aVTzSRpywNahzZBT695hMz81cVZJoZzaV0KaMTlSnBkrviPz1nIGHYCHJxF9r
 cIu0GSIyUjZ/7xslxdvjpLth16H27JCWDzDqIQMtg61063gNyEyWgt1qRSaK14JIH/DoYRfn
 jfFQSC8bffFjat7BQGFz4ZpRavkMUFuDirn5Tf28oc5ebe2cIHp4/kajTx/7JOxWZ80U70mA
 cBgEeYSrYYnX+UJsSxpzLc/0sT1eRJDEhI4XIQM4ClIzpsCIN5HnVF76UQXh3a9zpwh3dk8i
 bhN/URmCOTH+LHNJYN/MxY8wuukq877DWB7k86pBs5IDLAXmW8v3gIDWyIcgYqb2v8QO2Mqx
 YMqL7UZxVLul4/JbllsQB8F/fNI8AfttmAQL9cwo6C8yDTXKdho920W4WUR9k8NT/OBqWSyk
 bGqMHex48FVZhexNPYOd58EY9/7mL5u0sJmo+jTeb4JBgIbFPJCFyng4HwbniWgQJZ1WqaUC
 nas9J77uICis2WH7N8Bs9jy0wQYezNzqS+FxoNXmDQg2jetX8en4bO2Di7Pmx0jXA4TOb9TM
 izWDgYvmBE8=
In-Reply-To: <ZLeiM6l6tu6XDzrr@hovoldconsulting.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 19.07.2023 10:43, Johan Hovold wrote:
> On Tue, Jul 18, 2023 at 09:23:52AM -0700, Bjorn Andersson wrote:
>> On Tue, Jul 18, 2023 at 03:26:51PM +0200, Konrad Dybcio wrote:
>>> On 18.07.2023 15:20, Johan Hovold wrote:
>>>> On Mon, Jul 17, 2023 at 05:19:10PM +0200, Konrad Dybcio wrote:
>>>>> Some clocks need to be always-on, but we don't really do anything
>>>>> with them, other than calling enable() once and telling Linux they're
>>>>> enabled.
>>>>>
>>>>> Unregister them to save a couple of bytes and, perhaps more
>>>>> importantly, allow for runtime suspend of the clock controller device,
>>>>> as CLK_IS_CRITICAL prevents the latter.
>>>>
>>>> But this doesn't sound right. How can you disable a controller which
>>>> still has clocks enabled?
>>>>
>>>> Shouldn't instead these clocks be modelled properly so that they are
>>>> only enabled when actually needed?
>>> Hm.. We do have clk_branch2_aon_ops, but something still needs to
>>> toggle these clocks.
>>>
>>
>> Before we started replacing these clocks with static votes, I handled
>> exactly this problem in the turingcc-qcs404 driver by registering the
>> ahb clock with a pm_clk_add(). The clock framework will then
>> automagically keep the clock enabled around operations, but it will also
>> keep the runtime state active as long as the clock is prepared.
>>
>> As mentioned in an earlier reply today, there's no similarity to this in
>> the reset or gdsc code, so we'd need to add that in order to rely on
>> such mechanism.
> 
> This reminds me of:
> 
> 	4cc47e8add63 ("clk: qcom: gdsc: Remove direct runtime PM calls")
> 
> which recently removed a broken attempt to implement this for gdscs.
> 
> Just stumbled over GENPD_FLAG_PM_CLK which may provide a way forward at
> least for genpd (but see below).
> 
>>> I *think* we could leave a permanent vote in probe() without breaking
>>> runtime pm! I'll give it a spin bit later..
>>>
>>
>> Modelling the AHB clock in DT and putting a devm_clk_get_enabled() would
>> properly connect the two, and thereby handle probe order between the two
>> clock controllers.
> 
> Yeah, this dependency really should be described eventually.
> 
>> But it would prevent the power-domain of the parent provider to ever
>> suspending. Using pm_clk_add() this would at least depend on client
>> votes.
> 
> IIUC using pm_clk_add() would also prevent the parent from suspending
> due to that resume call in clk_prepare().
> 
> And this mechanism is also used for GENPD_FLAG_PM_CLK...
So.. how do we go about solving the issue that this patch tried to
address?

Konrad